An assertion utility that combines variable stop codes and error termination in pure
procedures to produce descriptive messages when a program detects violations of the requirements for correct execution.
- To mitigate against a reason developers often cite for not writing
pure
procedures: their inability to produce output in normal execution. - To promote the enforcement of programming contracts.
This assertion utility contains four public entities:
- An
assert
subroutine, - A
characterizable_t
abstract type supportingassert
, and - An
intrinsic_array_t
non-abstract type extendingcharacterizable_t
. - An
assert_macros.h
header file containing C-preprocessor macros.
The assert
subroutine
- Error-terminates with a variable stop code when a caller-provided logical assertion fails,
- Includes user-supplied diagnostic data in the output if provided by the calling procedure,
- Is callable inside
pure
procedures, and - Can be eliminated at compile-time, as controlled by the
ASSERTIONS
preprocessor define.
Assertion enforcement is controlled via the ASSERTIONS
preprocessor macro,
which can be defined to non-zero or zero at compilation time to
respectively enable or disable run-time assertion enforcement.
When the ASSERTIONS
preprocessor macro is not defined to any value,
the default is that assertions are disabled and will not check the condition.
To enable assertion enforcement (e.g., for a debug build), define the preprocessor ASSERTIONS to non-zero, eg:
fpm build --flag "-DASSERTIONS"
The program example/invoke-via-macro.F90 demonstrates the preferred way to invoke assertions via the three provided macros.
Invoking assertions this way ensures such calls will be completely removed whenever the ASSERTIONS
macro is undefined (or defined to zero) during compilation.
Due to a limitation of fpm
, this approach works best if the project using Assert is also a fpm
project.
If instead fpm install
is used, then either the user must copy include/assert_macros.h
to the installation directory (default: ~/.local/include
) or
the user must invoke assert
directly (via call assert(...)
).
In the latter approach when the assertions are disabled, the assert
procedure will start and end with if (.false.) then ... end if
, which might facilitate automatic removal of assert
during the dead-code removal phase of optimizing compilers.
The characterizable_t
type defines an as_character()
deferred binding that produces character
strings for use as diagnostic output from a user-defined derived type that extends characterizable_t
and implements the deferred binding.
The intrinsic_array_t
type that extends characterizable_t
provides a convenient mechanism for producing diagnostic output from arrays of intrinsic type complex
, integer
, logical
, or real
.
Two common use cases include
- Enforcing programming contracts throughout a project via runtime checks.
- Producing output in
pure
procedures for debugging purposes.
Programming can be thought of as requirements for correct execution of a procedure and assurances for the result of correct execution. The requirements and assurances might be constraints of three kinds:
- Preconditions (requirements):
logical
expressions that must evaluate to.true.
when a procedure starts execution, - Postconditions (assurances): expressions that must evaluate to
.true.
when a procedure finishes execution, and - Invariants: universal pre- and postconditions that must always be true when all procedures in a class start or finish executing.
The example/README.md file shows examples of writing constraints in notes on class diagrams using the formal syntax of the Object Constraint Language (OCL).
git clone [email protected]:berkeleylab/assert
cd assert
The following command builds Assert and runs the full test suite in a single image:
fpm test --profile release --flag "-ffree-line-length-0"
which builds the Assert library (with the default of assertion enforcement disabled) and runs the test suite.
With gfortran
and OpenCoarrays installed,
fpm test --compiler caf --profile release --runner "cafrun -n 2" --flag "-ffree-line-length-0"
To build and test with the Numerical Algorithms Group (NAG) Fortran compiler version 7.1 or later, use
fpm test --compiler=nagfor --profile release --flag "-coarray=cosmp -fpp -f2018"
fpm test --compiler ifx --profile release --flag -coarray
fpm test --compiler flang-new --flag "-mmlir -allow-assumed-rank -O3"
fpm test --compiler nagfor --profile release --flag "-fpp -coarray=cosmp"
Because fpm
uses the compiler name to determine the compiler identity and because
CCE provides one compiler wrapper, ftn
, for invoking all compilers, you will
need to invoke ftn
in a shell script named to identify CCE compiler. For example,
place a script named crayftn.sh
in your path with the following contents and with
executable privileges set appropriately:
#!/bin/bash
ftn $@
Then build and test Assert with the command
fpm test --compiler crayftn.sh --profile release
To use Assert with other compilers, please submit an issue or pull request.
See the ./example subdirectory.
See Assert's GitHub Pages site for HTML documentation generated with ford
.
For further documentation, please see example/README.md and the tests. Also, the code in src has comments formatted for generating HTML documentation using FORD.
The call_assert*
macros from the assert_macros.h
header file provide the
attractive guarantee that they will always compile completely away when
assertions are disabled, regardless of compiler analyses and optimization
level. This means users can reap the maintainability and correctness benefits
of aggressively asserting invariants throughout their code, without needing to
balance any potential performance cost associated with such assertions when the
code runs in production.
Unfortunately, C-preprocessor macros do not integrate cleanly with some aspects of the Fortran language. As such, you might encounter one or more of the following pitfalls when using these macros.
Up to and including the Fortran 2018 language standard, compilers were only
required to support up to 132 characters per free-form source line.
Preprocessor macro invocations are always expanded to a single line during
compilation, so when passing non-trivial arguments to macros including
call_assert*
it becomes easy for the expansion to exceed this line length
limit. This can result in compile-time errors like the following from gfortran:
Error: Line truncated at (1) [-Werror=line-truncation]
Some compilers offer a command-line argument that can be used to workaround this legacy limit, eg:
gfortran -ffree-line-length-0
akagfortran -ffree-line-length-none
When using fpm
, one can pass such a flag to the compiler using the fpm --flag
option, eg:
$ fpm test --profile release --flag -ffree-line-length-0
Thankfully Fortran 2023 raised this obscolecent line limit to 10,000 characters, so by using newer compilers you might never encounter this problem. In the case of gfortran, this appears to have been resolved by default starting in release 14.1.0.
The preprocessor is not currently specified by any Fortran standard, and as of 2024 its operation differs in subtle ways between compilers. One way in which compilers differ is how macro invocations can safely be broken across multiple lines.
For example, gfortran and flang-new both accept backslash \
continuation
character for line-breaks in a macro invocation:
! OK for flang-new and gfortran
call_assert_diagnose( computed_checksum == expected_checksum, \
"Checksum mismatch failure!", \
expected_checksum )
Whereas Cray Fortran wants &
line continuation characters, even inside
a macro invocation:
! OK for Cray Fortran
call_assert_diagnose( computed_checksum == expected_checksum, &
"Checksum mismatch failure!", &
expected_checksum )
There appears to be no syntax acceptable to all compilers, so when writing portable code it's probably best to avoid line breaks inside a macro invocation.
Fortran does not support comments with an end delimiter, only to-end-of-line comments. As such, there is no portable way to safely insert a Fortran comment into the middle of a macro invocation. For example, the following seemingly reasonable code results in a syntax error after macro expansion (on gfortran and flang-new):
! INCORRECT: cannot use Fortran comments inside macro invocation
call_assert_diagnose( computed_checksum == expected_checksum, ! ensured since version 3.14
"Checksum mismatch failure!", ! TODO: write a better message here
computed_checksum )
Depending on your compiler it might be possible to use a C-style block comment (because they are often removed by the preprocessor), for example with gfortran one can instead write the following:
call_assert_diagnose( computed_checksum == expected_checksum, /* ensured since version 3.14 */ \
"Checksum mismatch failure!", /* TODO: write a better message here */ \
computed_checksum )
However that capability might not be portable to other Fortran compilers. When in doubt, one can always move the comment outside the macro invocation:
! assert a property ensured since version 3.14
call_assert_diagnose( computed_checksum == expected_checksum, \
"Checksum mismatch failure!", \
computed_checksum ) ! TODO: write a better message above
See the LICENSE file for copyright and licensing information.